Syntactic Segmentation of Spontaneous Speech: Psychological and Cognitive Aspects

• Published in: Speech and Computer Vol. 11658; pp. 459 - 470
• Main Authors: Stepikhov, Anton, Loukina, Anastassia, Stepikhova, Natella
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 01.01.2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Annotation; Dominant hemisphere; Lateral asymmetry; Personality; Russian; Segmentation; Sentence boundary detection; Spontaneous speech; Working memory capacity
• ISBN: 3030260607; 9783030260606
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-26061-3_47

The paper examines the properties of expert manual annotation of Russian spontaneous speech. While it is well known that experts exhibit variability in the ways they mark transcripted speech, our aim is to arrive at the reasons behind such variability. In this study we focus on the annotator’s psychological profile (personality traits, working memory capacity, processing speed and lateral asymmetry). Our focus is to determine whether there is a relationship between the annotated sentence length and the psychological and cognitive characteristics of the annotator. We also study inter-annotator agreement in different text types. The participants (n = 80) detected sentence boundaries in the transcripts of Russian spontaneous speech and performed several test tasks. Personality traits were measured using the Five Factor Personality Inventory. Working memory capacity was measured through reading span and operation span tasks. To compute processing speed we used Letter Comparison and Pattern Comparison tasks. A dominant hemisphere for speech processing was established based on a dichotic listening task. The data analysis did not reveal any relationship between annotators individual characteristics and segmentation results. However, we found that annotators do tend to mark sentence length in a way that is individual to them and that such practices remain relatively stable regardless of text type or even language.